[U-Boot] [PATCH] arm: mvebu: move i2c slave disable to generic SPL code

[Maxime Ripard]

On Tue, May 29, 2018 at 05:43:30PM +0200, Stefan Roese wrote:
> > On Mon, May 28, 2018 at 11:04:38AM +0200, Stefan Roese wrote:
> > > (Added Jagan and Maxime to Cc)
> > > 
> > > On 28.05.2018 10:55, Chris Packham wrote:
> > > > Hi Baruch,
> > > > 
> > > > On Mon, May 28, 2018 at 8:22 PM Baruch Siach <baruch at tkos.co.il> wrote:
> > > > 
> > > > > Hi Chris,
> > > > 
> > > > > On Mon, May 28, 2018 at 08:11:39PM +1200, Chris Packham wrote:
> > > > > > On Mon, May 28, 2018 at 3:27 AM Baruch Siach <baruch at tkos.co.il> wrote:
> > > > > > > The hidden i2c slave that interferes the i2c bus is not board
> > > > specific.
> > > > > > > All Armada 38x SoCs are affected. Move the code disabling this slave
> > > > to
> > > > > > > generic code to make it work on all affected hardware.
> > > > > > 
> > > > > > I can't find a definition of this but the register seems to work for
> > > > > > kirkwood as well (not surprising since it's probably a common IP
> > > > block). Is
> > > > > > there any chance we can find a home for this that's available to boards
> > > > > > that don't use SPL?
> > > > 
> > > > > This workaround is Armada 38x specific. Are you aware of any 38x SoC based
> > > > > system that does not use SPL?
> > > > 
> > > > > As far as I can see 38x support requires SPL. CONFIG_ARMADA_38X selects
> > > > > CONFIG_ARMADA_32BIT that in turn selects CONFIG_SPL_DM. CONFIG_SPL_DM
> > > > depends
> > > > > on CONFIG_SPL.
> > > > 
> > > > The original workaround was implemented for the Turris Omnia board, as
> > > > you've highlighted this is debug register is documented for Armada-385. On
> > > > a hunch I tried clearing bit 18 of register 0xf101108c on a Kirkwood based
> > > > board and found that it also works to suppress the 0x64 self address. It's
> > > > highly likely that the i2c block is common to many Marvell SoCs and it just
> > > > happens to be documented on A-385.
> > > 
> > > Thanks for testing this Chris. This I2C IP core is also used by many
> > > Allwinner SoCs:
> > > 
> > > 	{ .compatible = "allwinner,sun6i-a31-i2c", },
> > > 
> > > 
> > > so it might be possible, that this "feature" is also available here.
> > > Perhaps somebody from the "sunxi front" can confirm this?
> > 
> > I went back to the thread a bit, but I couldn't really understand what
> > the issue was exactly. Do you have a spurious slave with the
> > controller at the address 0x64? When would it happen? Anytime you for
> > example probe the i2c bus?
> 
> On the Marvell platforms (Kirkwood, Armada 38x, possibly others), the
> SoC has the address 0x64 reserved as its I2C slave address (for other
> masters). So this address is not available for I2C devices attached
> to the I2C bus. With this patch (via this debug register), this slave
> address can be "free'ed" making it available for other I2C devices.

Ok. And that fantom slave can be accessed from the controller itself
when it's the master, or it's only usable by other masters on the bus?

ie, an i2cdetect running on that bus from that controller would detect
the slave you were mentionning? If so, then we don't have it on the
Allwinner SoCs (or it's not enabled by default at least).

Maxime

-- 
Maxime Ripard, Bootlin (formerly Free Electrons)
Embedded Linux and Kernel engineering
https://bootlin.com